Notes from FAST 2026
I attended the File and Storage Technologies (FAST) conference in Santa Clara last week. Some observations and thoughts:
"Art is never finished, only abandoned"
The keynote talk had a section emphasizing the importance of benchmarking the various components of the system (e.g., disk, network, CPU, memory, context switching, ...) to help understand the result of application performance. The point was well received but it did generate some interesting follow up conversations about how many (and how deep) rabbit holes should one pursue. After all, time is a limited resources and there might be diminishing returns. The speaker alluded to making a judgement call on what is good enough, citing the quote "art is never finished, only abandoned" often attributed to Leonardo DaVinci.
Quality of papers
FAST 2026 had a 17% acceptance rate and some of the papers were really intriguing. One that stood out (and also one of the best paper award winners) was SpecFS -- a file system generated purely by coding agents based off of human generated spec. I am generally an AI skeptic and a reluctant adopter. But I also understand that it is indeed a powerful tool and this was an interesting use case. Two other papers, MOST and HARE, were quite impressive in the simplicity and cleverness of their core idea. On the flip side, I was a bit disappointed to see some papers which, in my (trying-to-be-humble-but-failing) opinion did not quite meet the bar in terms of understanding and/or advancing the state of the art.
Unexpected attendees
Most of the conference attendees were the usual suspects, the professors, students, and (big-)tech workers. However, I was pleasantly surprised to meet a couple of folks who were trying to decide whether to pursue a systems PhD or not and sampling systems conferences like FAST to get a feel of research. There was another interesting attendee that I ran into: a developer from Europe visiting the US to attend a random sampling of developer and research conferences.
Networking is a long game
Each year at FAST, the Parallel Data Lab holds its reunion. I was a part of PDL during my PhD from 2015 to 2020 and have been to the reunion multiple time (2017, 2019, 2022, 2023, 2025, and 2026). In the earlier years, I did not recognize many of the attendees. Over the years, I have come to know (and become known) to many of the repeat attendees. I can now count many of the folks I have met through these reunions as part of my "network" -- folks I would not feel awkward reaching out to if/when I need some help. This is great, but it did not happen in a single shot. It took repetition. Multiple meetings and conversations without any specific agenda.